Social Media Mining

Case Study : Event Detection in TwitterBased on 'Event Detection in Twitter' by Jianshu Weng, Bu-Sung Lee ICWSM 2011

Presented by :

Yue HE & Falitokiniaina RABEARISON

June 2014

Motivation

Case Study : Event Detection in Twitter - 01

Outline

Related work

Presentation of EDCoW

Experiments and Results

Conclusion

Discussion

Case Study : Event Detection in Twitter - 03

Related WorkCase Study : Event Detection in Twitter - 05

Related Work

Term-weighting-based approaches

Topic-modeling-based approaches

Clustering-based approaches

Case Study : Event Detection in Twitter - 06

PRESENTATION of EDCoWCase Study : Event Detection in Twitter - 07

EDCoW [ Workflow]

Case Study : Event Detection in Twitter - 08

Case Study : Event Detection in Twitter - 09

EDCoW [Components]

EXPERIMENTS and RESULTSCase Study : Event Detection in Twitter - 10

Dataset Description and Preparation

Case Study : Event Detection in Twitter - 11

Signal Construction

Case Study : Event Detection in Twitter - 12

“santa + wine”

Case Study : Event Detection in Twitter - 13

Signal Construction

Case Study : Event Detection in Twitter - 14

Q: How can you group the sets of words?

A: Similarity between words

Auto Correlation Computation

2400 words 32 keywordsFilter: Median Absolute Deviation(MAD) :

Case Study : Event Detection in Twitter - 15

Cross Correlation Computation

Case Study : Event Detection in Twitter - 16

Cross Correlation Computation

Case Study : Event Detection in Twitter - 17

22 nodes 19 edges32 nodes 465 edges

(nodes=signals/keywords)

Reduction

Case Study : Event Detection in Twitter - 18

Modularity-based Graph Partition

Network contains 22 nodes and 19 edges.

After using Newman Alg, we find 6 clusters(event).

The events’ significance is presented by weights of

the edges between keywords and the lengths of the

event.

Case Study : Event Detection in Twitter - 19

Event

Case Study : Event Detection in Twitter - 20

Detect Event

Case Study : Event Detection in Twitter - 21

Detect Event

Case Study : Event Detection in Twitter - 22

CONCLUSIONCase Study : Event Detection in Twitter - 23

Conclusion

Case Study : Event Detection in Twitter - 24

● After studying and implementing the Event Detection with Clustering of

Wavelet-based Signals Algorithm in Java, we find it works in different time

period. We also try to use different parameters to evaluate it on the twitter

dataset.

● Some places to improve:

● How to control the parameters?

● How to let the keywords(events) make sense?

● It need some background to translate the combined keywords into event.

● How to utilize the results from public opinion?

Integrate EDCoW into SONDY

Case Study : Event Detection in Twitter - 25

EDCoW [Javadoc]

Case Study : Event Detection in Twitter - 26

Case Study : Event Detection in Twitter - 27

Acknowledge

Thank Jairo Cugliari & Adrien Guille for your

guides, comments and discussions!

Case Study : Event Detection in Twitter - 28

DISCUSSION

Thanks for your attention! ;)

Case Study : Event Detection in Twitter - 29